ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libeio/eio.c
(Generate patch)

Comparing libeio/eio.c (file contents):
Revision 1.11 by root, Tue May 13 18:50:27 2008 UTC vs.
Revision 1.150 by root, Fri Jun 15 02:57:36 2018 UTC

1/*
2 * libeio implementation
3 *
4 * Copyright (c) 2007,2008,2009,2010,2011,2012,2013,2016,2017 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met:
9 *
10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer.
12 *
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
22 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
23 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
24 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
25 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
26 * OF THE POSSIBILITY OF SUCH DAMAGE.
27 *
28 * Alternatively, the contents of this file may be used under the terms of
29 * the GNU General Public License ("GPL") version 2 or any later version,
30 * in which case the provisions of the GPL are applicable instead of
31 * the above. If you wish to allow the use of your version of this file
32 * only under the terms of the GPL and not to allow others to use your
33 * version of this file under the BSD license, indicate your decision
34 * by deleting the provisions above and replace them with the notice
35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL.
38 */
39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
1#include "eio.h" 44#include "eio.h"
2#include "xthread.h" 45#include "ecb.h"
3 46
4#include <errno.h> 47#include <errno.h>
5#include <stddef.h> 48#include <stddef.h>
6#include <stdlib.h> 49#include <stdlib.h>
7#include <string.h> 50#include <string.h>
… …
10#include <sys/stat.h> 53#include <sys/stat.h>
11#include <limits.h> 54#include <limits.h>
12#include <fcntl.h> 55#include <fcntl.h>
13#include <assert.h> 56#include <assert.h>
14 57
58/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
59/* intptr_t only comes from stdint.h, says idiot openbsd coder */
60#if HAVE_STDINT_H
61# include <stdint.h>
62#endif
63
64#ifndef ECANCELED
65# define ECANCELED EDOM
66#endif
67#ifndef ELOOP
68# define ELOOP EDOM
69#endif
70
71#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
72# define ENOTSOCK WSAENOTSOCK
73#endif
74
75static void eio_destroy (eio_req *req);
76
15#ifndef EIO_FINISH 77#ifndef EIO_FINISH
16# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 78# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
17#endif 79#endif
18 80
19#ifndef EIO_DESTROY 81#ifndef EIO_DESTROY
… …
22 84
23#ifndef EIO_FEED 85#ifndef EIO_FEED
24# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 86# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
25#endif 87#endif
26 88
89#ifndef EIO_FD_TO_WIN32_HANDLE
90# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
91#endif
92#ifndef EIO_WIN32_HANDLE_TO_FD
93# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
94#endif
95
96#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
97
98#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
99
27#ifdef _WIN32 100#ifdef _WIN32
28 101
29 /*doh*/ 102 #undef PAGESIZE
103 #define PAGESIZE 4096 /* GetSystemInfo? */
104
105 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
106
107 #ifdef EIO_STRUCT_STATI64
108 /* look at perl's non-sloppy stat */
109 #define stat(path,buf) _stati64 (path,buf)
110 #define fstat(fd,buf) _fstati64 (fd,buf)
111 #endif
112 #define lstat(path,buf) stat (path,buf)
113 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
114 #define mkdir(path,mode) _mkdir (path)
115 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
116
117 #define chmod(path,mode) _chmod (path, mode)
118 #define dup(fd) _dup (fd)
119 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
120 #define pipe(fds) _pipe (fds, 4096, O_BINARY)
121
122 #define fcntl(fd,cmd,arg) EIO_ENOSYS ()
123 #define ioctl(fd,cmd,arg) EIO_ENOSYS ()
124 #define fchmod(fd,mode) EIO_ENOSYS ()
125 #define chown(path,uid,gid) EIO_ENOSYS ()
126 #define fchown(fd,uid,gid) EIO_ENOSYS ()
127 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
128 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
129 #define mknod(path,mode,dev) EIO_ENOSYS ()
130 #define sync() EIO_ENOSYS ()
131 #define readlink(path,buf,s) EIO_ENOSYS ()
132 #define statvfs(path,buf) EIO_ENOSYS ()
133 #define fstatvfs(fd,buf) EIO_ENOSYS ()
134
135 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
136 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
137
138 #if __GNUC__
139 typedef long long eio_off_t; /* signed for compatibility to msvc */
140 #else
141 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
142 #endif
143
144 static eio_ssize_t
145 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
146 {
147 OVERLAPPED o = { 0 };
148 DWORD got;
149
150 o.Offset = offset;
151 o.OffsetHigh = offset >> 32;
152
153 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
154 ? got : -1;
155 }
156
157 static eio_ssize_t
158 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
159 {
160 OVERLAPPED o = { 0 };
161 DWORD got;
162
163 o.Offset = offset;
164 o.OffsetHigh = offset >> 32;
165
166 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
167 ? got : -1;
168 }
169
170 /* rename() uses MoveFile, which fails to overwrite */
171 #define rename(old,neu) eio__rename (old, neu)
172
173 static int
174 eio__rename (const char *old, const char *neu)
175 {
176 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
177 return 0;
178
179 /* should steal _dosmaperr */
180 switch (GetLastError ())
181 {
182 case ERROR_FILE_NOT_FOUND:
183 case ERROR_PATH_NOT_FOUND:
184 case ERROR_INVALID_DRIVE:
185 case ERROR_NO_MORE_FILES:
186 case ERROR_BAD_NETPATH:
187 case ERROR_BAD_NET_NAME:
188 case ERROR_BAD_PATHNAME:
189 case ERROR_FILENAME_EXCED_RANGE:
190 errno = ENOENT;
191 break;
192
193 default:
194 errno = EACCES;
195 break;
196 }
197
198 return -1;
199 }
200
201 /* we could even stat and see if it exists */
202 static int
203 symlink (const char *old, const char *neu)
204 {
205 #if WINVER >= 0x0600
206 int flags;
207
208 /* This tries out all combinations of SYMBOLIC_LINK_FLAG_DIRECTORY
209 * and SYMBOLIC_LINK_FLAG_ALLOW_UNPRIVILEGED_CREATE,
210 * with directory first.
211 */
212 for (flags = 3; flags >= 0; --flags)
213 if (CreateSymbolicLink (neu, old, flags))
214 return 0;
215 #endif
216
217 return EIO_ERRNO (ENOENT, -1);
218 }
219
220 /* POSIX API only, causing trouble for win32 apps */
221 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
222 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
223
224 struct statvfs
225 {
226 int dummy;
227 };
228
229 #define DT_DIR EIO_DT_DIR
230 #define DT_REG EIO_DT_REG
231 #define D_NAME(entp) entp.cFileName
232 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
30 233
31#else 234#else
32 235
33# include "config.h" 236 #include <sys/ioctl.h>
34# include <sys/time.h> 237 #include <sys/time.h>
35# include <sys/select.h> 238 #include <sys/select.h>
36# include <unistd.h> 239 #include <unistd.h>
240 #include <signal.h>
241 #include <dirent.h>
242
243 #ifdef ANDROID
244 #include <sys/vfs.h>
245 #define statvfs statfs
246 #define fstatvfs fstatfs
247 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
248 #else
249 #include <sys/statvfs.h>
250 #endif
251
252 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
253 #include <sys/mman.h>
254 #endif
255
256 #define D_NAME(entp) entp->d_name
257
258 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
259 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
260 #define _DIRENT_HAVE_D_TYPE /* sigh */
261 #define D_INO(de) (de)->d_fileno
262 #define D_NAMLEN(de) (de)->d_namlen
263 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
264 #define D_INO(de) (de)->d_ino
265 #endif
266
267 #ifdef _D_EXACT_NAMLEN
268 #undef D_NAMLEN
269 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
270 #endif
271
272 #ifdef _DIRENT_HAVE_D_TYPE
273 #define D_TYPE(de) (de)->d_type
274 #endif
275
276 #ifndef EIO_STRUCT_DIRENT
277 #define EIO_STRUCT_DIRENT struct dirent
278 #endif
279
280#endif
281
282#if HAVE_UTIMES
37# include <utime.h> 283# include <utime.h>
38# include <signal.h>
39# include <dirent.h>
40
41# ifndef EIO_STRUCT_DIRENT
42# define EIO_STRUCT_DIRENT struct dirent
43# endif 284#endif
44 285
286#if HAVE_SYS_SYSCALL_H
287# include <sys/syscall.h>
45#endif 288#endif
46 289
47#if HAVE_SENDFILE 290#if HAVE_SENDFILE
48# if __linux 291# if __linux
49# include <sys/sendfile.h> 292# include <sys/sendfile.h>
50# elif __freebsd 293# elif __FreeBSD__ || defined __APPLE__
51# include <sys/socket.h> 294# include <sys/socket.h>
52# include <sys/uio.h> 295# include <sys/uio.h>
53# elif __hpux 296# elif __hpux
54# include <sys/socket.h> 297# include <sys/socket.h>
55# elif __solaris /* not yet */ 298# elif __solaris
56# include <sys/sendfile.h> 299# include <sys/sendfile.h>
57# else 300# else
58# error sendfile support requested but not available 301# error sendfile support requested but not available
59# endif 302# endif
60#endif 303#endif
61 304
62/* number of seconds after which an idle threads exit */ 305#if HAVE_RENAMEAT2
63#define IDLE_TIMEOUT 10 306# include <sys/syscall.h>
307# include <linux/fs.h>
308#endif
309
310#ifndef D_TYPE
311# define D_TYPE(de) 0
312#endif
313#ifndef D_INO
314# define D_INO(de) 0
315#endif
316#ifndef D_NAMLEN
317# define D_NAMLEN(entp) strlen (D_NAME (entp))
318#endif
64 319
65/* used for struct dirent, AIX doesn't provide it */ 320/* used for struct dirent, AIX doesn't provide it */
66#ifndef NAME_MAX 321#ifndef NAME_MAX
67# define NAME_MAX 4096 322# define NAME_MAX 4096
68#endif 323#endif
69 324
325/* used for readlink etc. */
326#ifndef PATH_MAX
327# define PATH_MAX 0
328#endif
329
330#ifndef O_CLOEXEC
331 #define O_CLOEXEC 0
332#endif
333
334#ifndef EIO_PATH_MIN
335# define EIO_PATH_MIN 8160
336#endif
337
338#define EIO_PATH_MAX (PATH_MAX <= EIO_PATH_MIN ? EIO_PATH_MIN : PATH_MAX)
339
70/* buffer size for various temporary buffers */ 340/* buffer size for various temporary buffers */
71#define EIO_BUFSIZE 65536 341#define EIO_BUFSIZE 65536
72 342
73#define dBUF \ 343#define dBUF \
74 char *eio_buf; \
75 X_LOCK (wrklock); \
76 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 344 char *eio_buf = malloc (EIO_BUFSIZE); \
77 X_UNLOCK (wrklock); \
78 errno = ENOMEM; \ 345 errno = ENOMEM; \
79 if (!eio_buf) \ 346 if (!eio_buf) \
80 return -1; 347 return -1
81 348
82#define EIO_TICKS ((1000000 + 1023) >> 10) 349#define FUBd \
350 free (eio_buf)
83 351
84static void (*want_poll_cb) (void); 352/*****************************************************************************/
85static void (*done_poll_cb) (void);
86 353
87static unsigned int max_poll_time = 0; 354struct etp_tmpbuf;
88static unsigned int max_poll_reqs = 0;
89 355
90/* calculcate time difference in ~1/EIO_TICKS of a second */ 356#if _POSIX_VERSION >= 200809L
91static int tvdiff (struct timeval *tv1, struct timeval *tv2) 357 #define HAVE_AT 1
92{ 358 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
93 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 359 #ifndef O_SEARCH
94 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 360 #define O_SEARCH O_RDONLY
95} 361 #endif
362#else
363 #define HAVE_AT 0
364 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
365#endif
96 366
97static unsigned int started, idle, wanted = 4; 367struct eio_pwd
98
99/* worker threads management */
100static mutex_t wrklock = X_MUTEX_INIT;
101
102typedef struct worker
103{ 368{
104 /* locked by wrklock */ 369#if HAVE_AT
105 struct worker *prev, *next; 370 int fd;
106 371#endif
107 thread_t tid;
108
109 /* locked by reslock, reqlock or wrklock */
110 eio_req *req; /* currently processed request */
111 void *dbuf;
112 DIR *dirp;
113} worker;
114
115static worker wrk_first = { &wrk_first, &wrk_first, 0 };
116
117static void worker_clear (worker *wrk)
118{
119 if (wrk->dirp)
120 {
121 closedir (wrk->dirp);
122 wrk->dirp = 0;
123 }
124
125 if (wrk->dbuf)
126 {
127 free (wrk->dbuf);
128 wrk->dbuf = 0;
129 }
130}
131
132static void worker_free (worker *wrk)
133{
134 wrk->next->prev = wrk->prev;
135 wrk->prev->next = wrk->next;
136
137 free (wrk);
138}
139
140static volatile unsigned int nreqs, nready, npending;
141static volatile unsigned int max_idle = 4;
142
143static mutex_t reslock = X_MUTEX_INIT;
144static mutex_t reqlock = X_MUTEX_INIT;
145static cond_t reqwait = X_COND_INIT;
146
147unsigned int eio_nreqs (void)
148{
149 return nreqs;
150}
151
152unsigned int eio_nready (void)
153{
154 unsigned int retval;
155
156 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
157 retval = nready;
158 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
159
160 return retval;
161}
162
163unsigned int eio_npending (void)
164{
165 unsigned int retval;
166
167 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
168 retval = npending;
169 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
170
171 return retval;
172}
173
174unsigned int eio_nthreads (void)
175{
176 unsigned int retval;
177
178 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
179 retval = started;
180 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
181
182 return retval;
183}
184
185/*
186 * a somewhat faster data structure might be nice, but
187 * with 8 priorities this actually needs <20 insns
188 * per shift, the most expensive operation.
189 */
190typedef struct {
191 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */
192 int size; 372 int len;
193} reqq; 373 char str[1]; /* actually, a 0-terminated canonical path */
374};
194 375
195static reqq req_queue; 376/*****************************************************************************/
196static reqq res_queue;
197 377
198static int reqq_push (reqq *q, eio_req *req) 378#define ETP_PRI_MIN EIO_PRI_MIN
199{ 379#define ETP_PRI_MAX EIO_PRI_MAX
200 int pri = req->pri;
201 req->next = 0;
202 380
203 if (q->qe[pri]) 381#define ETP_TYPE_QUIT -1
204 { 382#define ETP_TYPE_GROUP EIO_GROUP
205 q->qe[pri]->next = req;
206 q->qe[pri] = req;
207 }
208 else
209 q->qe[pri] = q->qs[pri] = req;
210 383
211 return q->size++; 384static void eio_nop_callback (void) { }
212} 385static void (*eio_want_poll_cb)(void) = eio_nop_callback;
386static void (*eio_done_poll_cb)(void) = eio_nop_callback;
213 387
214static eio_req *reqq_shift (reqq *q) 388#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
215{ 389#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
216 int pri;
217 390
218 if (!q->size) 391struct etp_worker;
219 return 0; 392#define ETP_REQ eio_req
393#define ETP_DESTROY(req) eio_destroy (req)
394static int eio_finish (eio_req *req);
395#define ETP_FINISH(req) eio_finish (req)
396static void eio_execute (struct etp_worker *self, eio_req *req);
397#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
220 398
221 --q->size; 399#include "etp.c"
222 400
223 for (pri = EIO_NUM_PRI; pri--; ) 401static struct etp_pool eio_pool;
224 { 402#define EIO_POOL (&eio_pool)
225 eio_req *req = q->qs[pri];
226 403
227 if (req) 404/*****************************************************************************/
228 {
229 if (!(q->qs[pri] = (eio_req *)req->next))
230 q->qe[pri] = 0;
231 405
232 return req; 406static void
233 }
234 }
235
236 abort ();
237}
238
239static void grp_try_feed (eio_req *grp) 407grp_try_feed (eio_req *grp)
240{ 408{
241 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 409 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
242 { 410 {
243 int old_len = grp->size; 411 grp->flags &= ~ETP_FLAG_GROUPADD;
244 412
245 EIO_FEED (grp); 413 EIO_FEED (grp);
246 414
247 /* stop if no progress has been made */ 415 /* stop if no progress has been made */
248 if (old_len == grp->size) 416 if (!(grp->flags & ETP_FLAG_GROUPADD))
249 { 417 {
250 grp->feed = 0; 418 grp->feed = 0;
251 break; 419 break;
252 } 420 }
253 } 421 }
254} 422}
255 423
256static int eio_finish (eio_req *req); 424static int
257
258static int grp_dec (eio_req *grp) 425grp_dec (eio_req *grp)
259{ 426{
260 --grp->size; 427 --grp->size;
261 428
262 /* call feeder, if applicable */ 429 /* call feeder, if applicable */
263 grp_try_feed (grp); 430 grp_try_feed (grp);
264 431
265 /* finish, if done */ 432 /* finish, if done */
266 if (!grp->size && grp->int1) 433 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
267 return eio_finish (grp); 434 return eio_finish (grp);
268 else 435 else
269 return 0; 436 return 0;
270} 437}
271 438
439static void
272void eio_destroy (eio_req *req) 440eio_destroy (eio_req *req)
273{ 441{
274 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 442 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
275 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 443 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
276 444
277 EIO_DESTROY (req); 445 EIO_DESTROY (req);
278} 446}
279 447
448static int
280static int eio_finish (eio_req *req) 449eio_finish (eio_req *req)
281{ 450{
282 int res = EIO_FINISH (req); 451 int res = EIO_FINISH (req);
283 452
284 if (req->grp) 453 if (req->grp)
285 { 454 {
… …
293 if (grp->grp_first == req) 462 if (grp->grp_first == req)
294 grp->grp_first = req->grp_next; 463 grp->grp_first = req->grp_next;
295 464
296 res2 = grp_dec (grp); 465 res2 = grp_dec (grp);
297 466
298 if (!res && res2) 467 if (!res)
299 res = res2; 468 res = res2;
300 } 469 }
301 470
302 eio_destroy (req); 471 eio_destroy (req);
303 472
304 return res; 473 return res;
305} 474}
306 475
476void
307void eio_grp_cancel (eio_req *grp) 477eio_grp_cancel (eio_req *grp)
308{ 478{
309 for (grp = grp->grp_first; grp; grp = grp->grp_next) 479 etp_grp_cancel (EIO_POOL, grp);
310 eio_cancel (grp);
311} 480}
312 481
482void
313void eio_cancel (eio_req *req) 483eio_cancel (eio_req *req)
314{ 484{
315 X_LOCK (wrklock); 485 etp_cancel (EIO_POOL, req);
316 req->flags |= EIO_FLAG_CANCELLED;
317 X_UNLOCK (wrklock);
318
319 eio_grp_cancel (req);
320} 486}
321 487
322X_THREAD_PROC (eio_proc); 488void
323
324static void start_thread (void)
325{
326 worker *wrk = calloc (1, sizeof (worker));
327
328 /*TODO*/
329 assert (("unable to allocate worker thread data", wrk));
330
331 X_LOCK (wrklock);
332
333 if (thread_create (&wrk->tid, eio_proc, (void *)wrk))
334 {
335 wrk->prev = &wrk_first;
336 wrk->next = wrk_first.next;
337 wrk_first.next->prev = wrk;
338 wrk_first.next = wrk;
339 ++started;
340 }
341 else
342 free (wrk);
343
344 X_UNLOCK (wrklock);
345}
346
347static void maybe_start_thread (void)
348{
349 if (eio_nthreads () >= wanted)
350 return;
351
352 /* todo: maybe use idle here, but might be less exact */
353 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ())
354 return;
355
356 start_thread ();
357}
358
359void eio_submit (eio_req *req) 489eio_submit (eio_req *req)
360{ 490{
361 req->pri += EIO_PRI_BIAS; 491 etp_submit (EIO_POOL, req);
362
363 if (req->pri < EIO_PRI_MIN + EIO_PRI_BIAS) req->pri = EIO_PRI_MIN + EIO_PRI_BIAS;
364 if (req->pri > EIO_PRI_MAX + EIO_PRI_BIAS) req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;
365
366 ++nreqs;
367
368 X_LOCK (reqlock);
369 ++nready;
370 reqq_push (&req_queue, req);
371 X_COND_SIGNAL (reqwait);
372 X_UNLOCK (reqlock);
373
374 maybe_start_thread ();
375} 492}
376 493
377static void end_thread (void) 494unsigned int
495eio_nreqs (void)
378{ 496{
379 eio_req *req = calloc (1, sizeof (eio_req)); 497 return etp_nreqs (EIO_POOL);
380
381 req->type = EIO_QUIT;
382 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;
383
384 X_LOCK (reqlock);
385 reqq_push (&req_queue, req);
386 X_COND_SIGNAL (reqwait);
387 X_UNLOCK (reqlock);
388
389 X_LOCK (wrklock);
390 --started;
391 X_UNLOCK (wrklock);
392} 498}
393 499
500unsigned int
501eio_nready (void)
502{
503 return etp_nready (EIO_POOL);
504}
505
506unsigned int
507eio_npending (void)
508{
509 return etp_npending (EIO_POOL);
510}
511
512unsigned int ecb_cold
513eio_nthreads (void)
514{
515 return etp_nthreads (EIO_POOL);
516}
517
518void ecb_cold
394void eio_set_max_poll_time (double nseconds) 519eio_set_max_poll_time (double nseconds)
395{ 520{
396 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 521 etp_set_max_poll_time (EIO_POOL, nseconds);
397 max_poll_time = nseconds;
398 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
399} 522}
400 523
524void ecb_cold
401void eio_set_max_poll_reqs (unsigned int maxreqs) 525eio_set_max_poll_reqs (unsigned int maxreqs)
402{ 526{
403 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 527 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
404 max_poll_reqs = maxreqs;
405 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
406} 528}
407 529
530void ecb_cold
408void eio_set_max_idle (unsigned int nthreads) 531eio_set_max_idle (unsigned int nthreads)
409{ 532{
410 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 533 etp_set_max_idle (EIO_POOL, nthreads);
411 max_idle = nthreads <= 0 ? 1 : nthreads;
412 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
413} 534}
414 535
536void ecb_cold
537eio_set_idle_timeout (unsigned int seconds)
538{
539 etp_set_idle_timeout (EIO_POOL, seconds);
540}
541
542void ecb_cold
415void eio_set_min_parallel (unsigned int nthreads) 543eio_set_min_parallel (unsigned int nthreads)
416{ 544{
417 if (wanted < nthreads) 545 etp_set_min_parallel (EIO_POOL, nthreads);
418 wanted = nthreads;
419} 546}
420 547
548void ecb_cold
421void eio_set_max_parallel (unsigned int nthreads) 549eio_set_max_parallel (unsigned int nthreads)
422{ 550{
423 if (wanted > nthreads) 551 etp_set_max_parallel (EIO_POOL, nthreads);
424 wanted = nthreads;
425
426 while (started > wanted)
427 end_thread ();
428} 552}
429 553
430int eio_poll (void) 554int eio_poll (void)
431{ 555{
432 int maxreqs = max_poll_reqs; 556 return etp_poll (EIO_POOL);
433 struct timeval tv_start, tv_now;
434 eio_req *req;
435
436 if (max_poll_time)
437 gettimeofday (&tv_start, 0);
438
439 for (;;)
440 {
441 maybe_start_thread ();
442
443 X_LOCK (reslock);
444 req = reqq_shift (&res_queue);
445
446 if (req)
447 {
448 --npending;
449
450 if (!res_queue.size && done_poll_cb)
451 done_poll_cb ();
452 }
453
454 X_UNLOCK (reslock);
455
456 if (!req)
457 return 0;
458
459 --nreqs;
460
461 if (req->type == EIO_GROUP && req->size)
462 {
463 req->int1 = 1; /* mark request as delayed */
464 continue;
465 }
466 else
467 {
468 int res = eio_finish (req);
469 if (res)
470 return res;
471 }
472
473 if (maxreqs && !--maxreqs)
474 break;
475
476 if (max_poll_time)
477 {
478 gettimeofday (&tv_now, 0);
479
480 if (tvdiff (&tv_start, &tv_now) >= max_poll_time)
481 break;
482 }
483 }
484
485 errno = EAGAIN;
486 return -1;
487} 557}
488 558
489/*****************************************************************************/ 559/*****************************************************************************/
490/* work around various missing functions */ 560/* work around various missing functions */
491 561
492#if !HAVE_PREADWRITE 562#if HAVE_POSIX_CLOSE && !__linux
493# define pread eio__pread 563# define eio__close(fd) posix_close (fd, 0)
494# define pwrite eio__pwrite 564#else
495 565# define eio__close(fd) close (fd)
496/*
497 * make our pread/pwrite safe against themselves, but not against
498 * normal read/write by using a mutex. slows down execution a lot,
499 * but that's your problem, not mine.
500 */
501static mutex_t preadwritelock = X_MUTEX_INIT;
502
503static ssize_t
504eio__pread (int fd, void *buf, size_t count, off_t offset)
505{
506 ssize_t res;
507 off_t ooffset;
508
509 X_LOCK (preadwritelock);
510 ooffset = lseek (fd, 0, SEEK_CUR);
511 lseek (fd, offset, SEEK_SET);
512 res = read (fd, buf, count);
513 lseek (fd, ooffset, SEEK_SET);
514 X_UNLOCK (preadwritelock);
515
516 return res;
517}
518
519static ssize_t
520eio__pwrite (int fd, void *buf, size_t count, off_t offset)
521{
522 ssize_t res;
523 off_t ooffset;
524
525 X_LOCK (preadwritelock);
526 ooffset = lseek (fd, 0, SEEK_CUR);
527 lseek (fd, offset, SEEK_SET);
528 res = write (fd, buf, count);
529 lseek (fd, offset, SEEK_SET);
530 X_UNLOCK (preadwritelock);
531
532 return res;
533}
534#endif 566#endif
535 567
568/* close() without disturbing errno */
569static void
570silent_close (int fd)
571{
572 int saved_errno = errno;
573 eio__close (fd);
574 errno = saved_errno;
575}
576
536#ifndef HAVE_FUTIMES 577#ifndef HAVE_UTIMES
537 578
579# undef utimes
538# define utimes(path,times) eio__utimes (path, times) 580# define utimes(path,times) eio__utimes (path, times)
539# define futimes(fd,times) eio__futimes (fd, times)
540 581
541static int 582static int
542eio__utimes (const char *filename, const struct timeval times[2]) 583eio__utimes (const char *filename, const struct timeval times[2])
543{ 584{
544 if (times) 585 if (times)
… …
552 } 593 }
553 else 594 else
554 return utime (filename, 0); 595 return utime (filename, 0);
555} 596}
556 597
598#endif
599
600#ifndef HAVE_FUTIMES
601
602# undef futimes
603# define futimes(fd,times) eio__futimes (fd, times)
604
605static int
557static int eio__futimes (int fd, const struct timeval tv[2]) 606eio__futimes (int fd, const struct timeval tv[2])
558{ 607{
559 errno = ENOSYS; 608 errno = ENOSYS;
560 return -1; 609 return -1;
561} 610}
562 611
563#endif 612#endif
564 613
565#if !HAVE_FDATASYNC 614#if !HAVE_FDATASYNC
615# undef fdatasync
566# define fdatasync fsync 616# define fdatasync(fd) fsync (fd)
567#endif 617#endif
618
619static int
620eio__syncfs (int fd)
621{
622 int res;
623
624#if HAVE_SYS_SYNCFS
625 res = (int)syscall (__NR_syncfs, (int)(fd));
626#else
627 res = EIO_ENOSYS ();
628#endif
629
630 if (res < 0 && errno == ENOSYS && fd >= 0)
631 sync ();
632
633 return res;
634}
635
636/* sync_file_range always needs emulation */
637static int
638eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
639{
640#if HAVE_SYNC_FILE_RANGE
641 int res;
642
643 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
644 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
645 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
646 {
647 flags = 0
648 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
649 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
650 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
651 }
652
653 res = sync_file_range (fd, offset, nbytes, flags);
654
655 if (!res || errno != ENOSYS)
656 return res;
657#endif
658
659 /* even though we could play tricks with the flags, it's better to always
660 * call fdatasync, as that matches the expectation of its users best */
661 return fdatasync (fd);
662}
663
664static int
665eio__fallocate (int fd, int mode, off_t offset, size_t len)
666{
667#if HAVE_LINUX_FALLOCATE
668 return fallocate (fd, mode, offset, len);
669#else
670 return EIO_ENOSYS ();
671#endif
672}
568 673
569#if !HAVE_READAHEAD 674#if !HAVE_READAHEAD
675# undef readahead
570# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 676# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
571 677
572static ssize_t 678static eio_ssize_t
573eio__readahead (int fd, off_t offset, size_t count, worker *self) 679eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
574{ 680{
575 size_t todo = count; 681 size_t todo = count;
576 dBUF; 682 dBUF;
577 683
578 while (todo > 0) 684 while (todo > 0)
… …
582 pread (fd, eio_buf, len, offset); 688 pread (fd, eio_buf, len, offset);
583 offset += len; 689 offset += len;
584 todo -= len; 690 todo -= len;
585 } 691 }
586 692
587 errno = 0; 693 FUBd;
694
695 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
696 /* but not for e.g. EIO or eof, so we also never fail */
588 return count; 697 return 0;
589} 698}
590 699
591#endif
592
593#if !HAVE_READDIR_R
594# define readdir_r eio__readdir_r
595
596static mutex_t readdirlock = X_MUTEX_INIT;
597
598static int
599eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
600{
601 EIO_STRUCT_DIRENT *e;
602 int errorno;
603
604 X_LOCK (readdirlock);
605
606 e = readdir (dirp);
607 errorno = errno;
608
609 if (e)
610 {
611 *res = ent;
612 strcpy (ent->d_name, e->d_name);
613 }
614 else
615 *res = 0;
616
617 X_UNLOCK (readdirlock);
618
619 errno = errorno;
620 return e ? 0 : -1;
621}
622#endif 700#endif
623 701
624/* sendfile always needs emulation */ 702/* sendfile always needs emulation */
625static ssize_t 703static eio_ssize_t
626eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self) 704eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
627{ 705{
706 eio_ssize_t written = 0;
628 ssize_t res; 707 eio_ssize_t res;
629 708
630 if (!count) 709 if (!count)
631 return 0; 710 return 0;
632 711
712 for (;;)
713 {
714#ifdef __APPLE__
715# undef HAVE_SENDFILE /* broken, as everything on os x */
716#endif
633#if HAVE_SENDFILE 717#if HAVE_SENDFILE
634# if __linux 718# if __linux
719 off_t soffset = offset;
635 res = sendfile (ofd, ifd, &offset, count); 720 res = sendfile (ofd, ifd, &soffset, count);
636 721
637# elif __freebsd 722# elif __FreeBSD__
638 /* 723 /*
639 * Of course, the freebsd sendfile is a dire hack with no thoughts 724 * Of course, the freebsd sendfile is a dire hack with no thoughts
640 * wasted on making it similar to other I/O functions. 725 * wasted on making it similar to other I/O functions.
641 */ 726 */
642 {
643 off_t sbytes; 727 off_t sbytes;
644 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 728 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
645 729
646 if (res < 0 && sbytes) 730 #if 0 /* according to the manpage, this is correct, but broken behaviour */
647 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 731 /* freebsd' sendfile will return 0 on success */
732 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
733 /* not on e.g. EIO or EPIPE - sounds broken */
734 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
648 res = sbytes; 735 res = sbytes;
649 } 736 #endif
737
738 /* according to source inspection, this is correct, and useful behaviour */
739 if (sbytes)
740 res = sbytes;
741
742# elif defined __APPLE__
743 off_t sbytes = count;
744 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
745
746 /* according to the manpage, sbytes is always valid */
747 if (sbytes)
748 res = sbytes;
650 749
651# elif __hpux 750# elif __hpux
652 res = sendfile (ofd, ifd, offset, count, 0, 0); 751 res = sendfile (ofd, ifd, offset, count, 0, 0);
653 752
654# elif __solaris 753# elif __solaris
655 {
656 struct sendfilevec vec; 754 struct sendfilevec vec;
657 size_t sbytes; 755 size_t sbytes;
658 756
659 vec.sfv_fd = ifd; 757 vec.sfv_fd = ifd;
660 vec.sfv_flag = 0; 758 vec.sfv_flag = 0;
661 vec.sfv_off = offset; 759 vec.sfv_off = offset;
662 vec.sfv_len = count; 760 vec.sfv_len = count;
663 761
664 res = sendfilev (ofd, &vec, 1, &sbytes); 762 res = sendfilev (ofd, &vec, 1, &sbytes);
665 763
666 if (res < 0 && sbytes) 764 if (res < 0 && sbytes)
667 res = sbytes; 765 res = sbytes;
668 }
669 766
670# endif 767# endif
768
769#elif defined (_WIN32) && 0
770 /* does not work, just for documentation of what would need to be done */
771 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
772 /* libeio guarantees that the file offset does not change, and windows */
773 /* has no way to get an independent handle to the same file description */
774 HANDLE h = TO_SOCKET (ifd);
775 SetFilePointer (h, offset, 0, FILE_BEGIN);
776 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
777
671#else 778#else
672 res = -1; 779 res = EIO_ENOSYS ();
673 errno = ENOSYS;
674#endif 780#endif
675 781
782 /* we assume sendfile can copy at least 128mb in one go */
783 if (res <= 128 * 1024 * 1024)
784 {
785 if (res > 0)
786 written += res;
787
788 if (written)
789 return written;
790
791 break;
792 }
793 else
794 {
795 /* if we requested more, then probably the kernel was lazy */
796 written += res;
797 offset += res;
798 count -= res;
799
800 if (!count)
801 return written;
802 }
803 }
804
676 if (res < 0 805 if (res < 0
677 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 806 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
807 /* BSDs */
808#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
809 || errno == ENOTSUP
810#endif
811#ifdef EOPNOTSUPP /* windows */
812 || errno == EOPNOTSUPP /* BSDs */
813#endif
678#if __solaris 814#if __solaris
679 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 815 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
680#endif 816#endif
681 ) 817 )
682 ) 818 )
… …
686 822
687 res = 0; 823 res = 0;
688 824
689 while (count) 825 while (count)
690 { 826 {
691 ssize_t cnt; 827 eio_ssize_t cnt;
692 828
693 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 829 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
694 830
695 if (cnt <= 0) 831 if (cnt <= 0)
696 { 832 {
… …
708 844
709 offset += cnt; 845 offset += cnt;
710 res += cnt; 846 res += cnt;
711 count -= cnt; 847 count -= cnt;
712 } 848 }
849
850 FUBd;
713 } 851 }
714 852
715 return res; 853 return res;
854}
855
856#ifdef PAGESIZE
857# define eio_pagesize() PAGESIZE
858#else
859static intptr_t
860eio_pagesize (void)
861{
862 static intptr_t page;
863
864 if (!page)
865 page = sysconf (_SC_PAGESIZE);
866
867 return page;
868}
869#endif
870
871static void
872eio_page_align (void **addr, size_t *length)
873{
874 intptr_t mask = eio_pagesize () - 1;
875
876 /* round down addr */
877 intptr_t adj = mask & (intptr_t)*addr;
878
879 *addr = (void *)((intptr_t)*addr - adj);
880 *length += adj;
881
882 /* round up length */
883 *length = (*length + mask) & ~mask;
884}
885
886#if !_POSIX_MEMLOCK
887# define eio__mlockall(a) EIO_ENOSYS ()
888#else
889
890static int
891eio__mlockall (int flags)
892{
893 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
894 extern int mallopt (int, int);
895 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
896 #endif
897
898 if (EIO_MCL_CURRENT != MCL_CURRENT
899 || EIO_MCL_FUTURE != MCL_FUTURE)
900 {
901 flags = 0
902 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
903 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
904 }
905
906 return mlockall (flags);
907}
908#endif
909
910#if !_POSIX_MEMLOCK_RANGE
911# define eio__mlock(a,b) EIO_ENOSYS ()
912#else
913
914static int
915eio__mlock (void *addr, size_t length)
916{
917 eio_page_align (&addr, &length);
918
919 return mlock (addr, length);
920}
921
922#endif
923
924#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
925# define eio__msync(a,b,c) EIO_ENOSYS ()
926#else
927
928static int
929eio__msync (void *mem, size_t len, int flags)
930{
931 eio_page_align (&mem, &len);
932
933 if (EIO_MS_ASYNC != MS_SYNC
934 || EIO_MS_INVALIDATE != MS_INVALIDATE
935 || EIO_MS_SYNC != MS_SYNC)
936 {
937 flags = 0
938 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
939 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
940 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
941 }
942
943 return msync (mem, len, flags);
944}
945
946#endif
947
948static int
949eio__mtouch (eio_req *req)
950{
951 void *mem = req->ptr2;
952 size_t len = req->size;
953 int flags = req->int1;
954
955 eio_page_align (&mem, &len);
956
957 {
958 intptr_t addr = (intptr_t)mem;
959 intptr_t end = addr + len;
960 intptr_t page = eio_pagesize ();
961
962 if (addr < end)
963 if (flags & EIO_MT_MODIFY) /* modify */
964 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < end && !EIO_CANCELLED (req));
965 else
966 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < end && !EIO_CANCELLED (req));
967 }
968
969 return 0;
970}
971
972/*****************************************************************************/
973/* requests implemented outside eio_execute, because they are so large */
974
975static void
976eio__lseek (eio_req *req)
977{
978 /* this usually gets optimised away completely, or your compiler sucks, */
979 /* or the whence constants really are not 0, 1, 2 */
980 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
981 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
982 : req->int2 == EIO_SEEK_END ? SEEK_END
983 : req->int2;
984
985 req->offs = lseek (req->int1, req->offs, whence);
986 req->result = req->offs == (off_t)-1 ? -1 : 0;
987}
988
989/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
990static int
991eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
992{
993 char *res;
994 const char *rel = path;
995 char *tmp1, *tmp2;
996#if SYMLOOP_MAX > 32
997 int symlinks = SYMLOOP_MAX;
998#else
999 int symlinks = 32;
1000#endif
1001
1002 errno = EINVAL;
1003 if (!rel)
1004 return -1;
1005
1006 errno = ENOENT;
1007 if (!*rel)
1008 return -1;
1009
1010 res = etp_tmpbuf_get (tmpbuf, EIO_PATH_MAX * 3);
1011#ifdef _WIN32
1012 if (_access (rel, 4) != 0)
1013 return -1;
1014
1015 symlinks = GetFullPathName (rel, EIO_PATH_MAX * 3, res, 0);
1016
1017 errno = ENAMETOOLONG;
1018 if (symlinks >= EIO_PATH_MAX * 3)
1019 return -1;
1020
1021 errno = EIO;
1022 if (symlinks <= 0)
1023 return -1;
1024
1025 return symlinks;
1026
1027#else
1028 tmp1 = res + EIO_PATH_MAX;
1029 tmp2 = tmp1 + EIO_PATH_MAX;
1030
1031#if 0 /* disabled, the musl way to do things is just too racy */
1032#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1033 /* on linux we may be able to ask the kernel */
1034 {
1035 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1036
1037 if (fd >= 0)
1038 {
1039 sprintf (tmp1, "/proc/self/fd/%d", fd);
1040 req->result = readlink (tmp1, res, EIO_PATH_MAX);
1041 /* here we should probably stat the open file and the disk file, to make sure they still match */
1042 eio__close (fd);
1043
1044 if (req->result > 0)
1045 goto done;
1046 }
1047 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1048 return -1;
1049 }
1050#endif
1051#endif
1052
1053 if (*rel != '/')
1054 {
1055 int len;
1056
1057 errno = ENOENT;
1058 if (wd == EIO_INVALID_WD)
1059 return -1;
1060
1061 if (wd == EIO_CWD)
1062 {
1063 if (!getcwd (res, EIO_PATH_MAX))
1064 return -1;
1065
1066 len = strlen (res);
1067 }
1068 else
1069 memcpy (res, wd->str, len = wd->len);
1070
1071 if (res [1]) /* only use if not / */
1072 res += len;
1073 }
1074
1075 while (*rel)
1076 {
1077 eio_ssize_t len, linklen;
1078 const char *beg = rel;
1079
1080 while (*rel && *rel != '/')
1081 ++rel;
1082
1083 len = rel - beg;
1084
1085 if (!len) /* skip slashes */
1086 {
1087 ++rel;
1088 continue;
1089 }
1090
1091 if (beg [0] == '.')
1092 {
1093 if (len == 1)
1094 continue; /* . - nop */
1095
1096 if (beg [1] == '.' && len == 2)
1097 {
1098 /* .. - back up one component, if possible */
1099
1100 while (res != tmpbuf->ptr)
1101 if (*--res == '/')
1102 break;
1103
1104 continue;
1105 }
1106 }
1107
1108 errno = ENAMETOOLONG;
1109 if (res + 1 + len + 1 >= tmp1)
1110 return -1;
1111
1112 /* copy one component */
1113 *res = '/';
1114 memcpy (res + 1, beg, len);
1115
1116 /* zero-terminate, for readlink */
1117 res [len + 1] = 0;
1118
1119 /* now check if it's a symlink */
1120 linklen = readlink (tmpbuf->ptr, tmp1, EIO_PATH_MAX);
1121
1122 if (linklen < 0)
1123 {
1124 if (errno != EINVAL)
1125 return -1;
1126
1127 /* it's a normal directory. hopefully */
1128 res += len + 1;
1129 }
1130 else
1131 {
1132 /* yay, it was a symlink - build new path in tmp2 */
1133 int rellen = strlen (rel);
1134
1135 errno = ENAMETOOLONG;
1136 if (linklen + 1 + rellen >= EIO_PATH_MAX) /* also catch linklen >= EIO_PATH_MAX */
1137 return -1;
1138
1139 errno = ELOOP;
1140 if (!--symlinks)
1141 return -1;
1142
1143 if (*tmp1 == '/')
1144 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1145
1146 /* we need to be careful, as rel might point into tmp2 already */
1147 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1148 tmp2 [linklen] = '/';
1149 memcpy (tmp2, tmp1, linklen);
1150
1151 rel = tmp2;
1152 }
1153 }
1154
1155 /* special case for the lone root path */
1156 if (res == tmpbuf->ptr)
1157 *res++ = '/';
1158
1159 return res - (char *)tmpbuf->ptr;
1160#endif
1161}
1162
1163static signed char
1164eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1165{
1166 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1167 : a->inode < b->inode ? -1
1168 : a->inode > b->inode ? 1
1169 : 0;
1170}
1171
1172#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1173
1174#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1175#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1176
1177static void
1178eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1179{
1180 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1181 unsigned char *bit = bits;
1182
1183 assert (CHAR_BIT == 8);
1184 assert (sizeof (eio_dirent) * 8 < 256);
1185 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1186 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1187
1188 if (size <= EIO_SORT_FAST)
1189 return;
1190
1191 /* first prepare an array of bits to test in our radix sort */
1192 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1193 /* inode_bits must contain all inodes ORed together */
1194 /* which is used to skip bits that are 0 everywhere, which is very common */
1195 {
1196 eio_ino_t endianness;
1197 int i, j;
1198
1199 /* we store the byte offset of byte n into byte n of "endianness" */
1200 for (i = 0; i < sizeof (eio_ino_t); ++i)
1201 ((unsigned char *)&endianness)[i] = i;
1202
1203 *bit++ = 0;
1204
1205 for (i = 0; i < sizeof (eio_ino_t); ++i)
1206 {
1207 /* shifting off the byte offsets out of "endianness" */
1208 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1209 endianness >>= 8;
1210
1211 for (j = 0; j < 8; ++j)
1212 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1213 *bit++ = offs + j;
1214 }
1215
1216 for (j = 0; j < 8; ++j)
1217 if (score_bits & (1 << j))
1218 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1219 }
1220
1221 /* now actually do the sorting (a variant of MSD radix sort) */
1222 {
1223 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1224 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1225 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1226 int stk_idx = 0;
1227
1228 base_stk [stk_idx] = dents;
1229 end_stk [stk_idx] = dents + size;
1230 bit_stk [stk_idx] = bit - 1;
1231
1232 do
1233 {
1234 base = base_stk [stk_idx];
1235 end = end_stk [stk_idx];
1236 bit = bit_stk [stk_idx];
1237
1238 for (;;)
1239 {
1240 unsigned char O = *bit >> 3;
1241 unsigned char M = 1 << (*bit & 7);
1242
1243 eio_dirent *a = base;
1244 eio_dirent *b = end;
1245
1246 if (b - a < EIO_SORT_CUTOFF)
1247 break;
1248
1249 /* now bit-partition the array on the bit */
1250 /* this ugly asymmetric loop seems to perform much better than typical */
1251 /* partition algos found in the literature */
1252 do
1253 if (!(((unsigned char *)a)[O] & M))
1254 ++a;
1255 else if (!(((unsigned char *)--b)[O] & M))
1256 {
1257 eio_dirent tmp = *a; *a = *b; *b = tmp;
1258 ++a;
1259 }
1260 while (b > a);
1261
1262 /* next bit, or stop, if no bits left in this path */
1263 if (!*--bit)
1264 break;
1265
1266 base_stk [stk_idx] = a;
1267 end_stk [stk_idx] = end;
1268 bit_stk [stk_idx] = bit;
1269 ++stk_idx;
1270
1271 end = a;
1272 }
1273 }
1274 while (stk_idx--);
1275 }
1276}
1277
1278static void
1279eio_dent_insertion_sort (eio_dirent *dents, int size)
1280{
1281 /* first move the smallest element to the front, to act as a sentinel */
1282 {
1283 int i;
1284 eio_dirent *min = dents;
1285
1286 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1287 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1288 if (EIO_DENT_CMP (dents [i], <, *min))
1289 min = &dents [i];
1290
1291 /* swap elements 0 and j (minimum) */
1292 {
1293 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1294 }
1295 }
1296
1297 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1298 {
1299 eio_dirent *i, *j;
1300
1301 for (i = dents + 1; i < dents + size; ++i)
1302 {
1303 eio_dirent value = *i;
1304
1305 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1306 j [1] = j [0];
1307
1308 j [1] = value;
1309 }
1310 }
1311}
1312
1313static void
1314eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1315{
1316 if (size <= 1)
1317 return; /* our insertion sort relies on size > 0 */
1318
1319 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1320 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1321 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1322
1323 /* use an insertion sort at the end, or for small arrays, */
1324 /* as insertion sort is more efficient for small partitions */
1325 eio_dent_insertion_sort (dents, size);
716} 1326}
717 1327
718/* read a full directory */ 1328/* read a full directory */
719static void 1329static void
720eio__scandir (eio_req *req, worker *self) 1330eio__scandir (eio_req *req, etp_worker *self)
721{ 1331{
1332 char *name, *names;
1333 int namesalloc = 4096 - sizeof (void *) * 4;
1334 int namesoffs = 0;
1335 int flags = req->int1;
1336 eio_dirent *dents = 0;
1337 int dentalloc = 128;
1338 int dentoffs = 0;
1339 eio_ino_t inode_bits = 0;
1340#ifdef _WIN32
1341 HANDLE dirp;
1342 WIN32_FIND_DATA entp;
1343#else
722 DIR *dirp; 1344 DIR *dirp;
723 union
724 {
725 EIO_STRUCT_DIRENT d;
726 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
727 } *u;
728 EIO_STRUCT_DIRENT *entp; 1345 EIO_STRUCT_DIRENT *entp;
729 char *name, *names; 1346#endif
730 int memlen = 4096;
731 int memofs = 0;
732 int res = 0;
733 1347
734 X_LOCK (wrklock); 1348 req->result = -1;
735 self->dirp = dirp = opendir (req->ptr1);
736 self->dbuf = u = malloc (sizeof (*u));
737 req->flags |= EIO_FLAG_PTR2_FREE;
738 req->ptr2 = names = malloc (memlen);
739 X_UNLOCK (wrklock);
740 1349
741 if (dirp && u && names) 1350 if (!(flags & EIO_READDIR_DENTS))
742 for (;;) 1351 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1352
1353#ifdef _WIN32
1354 {
1355 int len = strlen ((const char *)req->ptr1);
1356 char *path = malloc (MAX_PATH);
1357 const char *fmt;
1358 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1359
1360 if (!len)
1361 fmt = "./*";
1362 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1363 fmt = "%s*";
1364 else
1365 fmt = "%s/*";
1366
1367 _snprintf (path, MAX_PATH, fmt, reqpath);
1368 dirp = FindFirstFile (path, &entp);
1369 free (path);
1370
1371 if (dirp == INVALID_HANDLE_VALUE)
1372 {
1373 /* should steal _dosmaperr */
1374 switch (GetLastError ())
1375 {
1376 case ERROR_FILE_NOT_FOUND:
1377 req->result = 0;
1378 break;
1379
1380 case ERROR_INVALID_NAME:
1381 case ERROR_PATH_NOT_FOUND:
1382 case ERROR_NO_MORE_FILES:
1383 errno = ENOENT;
1384 break;
1385
1386 case ERROR_NOT_ENOUGH_MEMORY:
1387 errno = ENOMEM;
1388 break;
1389
1390 default:
1391 errno = EINVAL;
1392 break;
1393 }
1394
1395 return;
1396 }
1397 }
1398#else
1399 #if HAVE_AT
1400 if (req->wd)
743 { 1401 {
1402 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1403
1404 if (fd < 0)
1405 return;
1406
1407 dirp = fdopendir (fd);
1408
1409 if (!dirp)
1410 silent_close (fd);
1411 }
1412 else
1413 dirp = opendir (req->ptr1);
1414 #else
1415 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1416 #endif
1417
1418 if (!dirp)
1419 return;
1420#endif
1421
1422 if (req->flags & EIO_FLAG_PTR1_FREE)
1423 free (req->ptr1);
1424
1425 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1426 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1427 req->ptr2 = names = malloc (namesalloc);
1428
1429 if (!names || (flags && !dents))
1430 return;
1431
1432 for (;;)
1433 {
1434 int done;
1435
1436#ifdef _WIN32
1437 done = !dirp;
1438#else
744 errno = 0; 1439 errno = 0;
745 readdir_r (dirp, &u->d, &entp); 1440 entp = readdir (dirp);
1441 done = !entp;
1442#endif
746 1443
1444 if (done)
1445 {
1446#ifndef _WIN32
1447 int old_errno = errno;
1448 closedir (dirp);
1449 errno = old_errno;
1450
747 if (!entp) 1451 if (errno)
1452 break;
1453#endif
1454
1455 /* sort etc. */
1456 req->int1 = flags;
1457 req->result = dentoffs;
1458
1459 if (flags & EIO_READDIR_STAT_ORDER)
1460 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1461 else if (flags & EIO_READDIR_DIRS_FIRST)
1462 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1463 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1464 else
1465 {
1466 /* in this case, all is known, and we just put dirs first and sort them */
1467 eio_dirent *oth = dents + dentoffs;
1468 eio_dirent *dir = dents;
1469
1470 /* now partition dirs to the front, and non-dirs to the back */
1471 /* by walking from both sides and swapping if necessary */
1472 while (oth > dir)
1473 {
1474 if (dir->type == EIO_DT_DIR)
1475 ++dir;
1476 else if ((--oth)->type == EIO_DT_DIR)
1477 {
1478 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1479
1480 ++dir;
1481 }
1482 }
1483
1484 /* now sort the dirs only (dirs all have the same score) */
1485 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1486 }
1487
748 break; 1488 break;
1489 }
749 1490
750 name = entp->d_name; 1491 /* now add the entry to our list(s) */
1492 name = D_NAME (entp);
751 1493
1494 /* skip . and .. entries */
752 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1495 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
753 { 1496 {
754 int len = strlen (name) + 1; 1497 int len = D_NAMLEN (entp) + 1;
755 1498
756 res++; 1499 while (ecb_expect_false (namesoffs + len > namesalloc))
757
758 while (memofs + len > memlen)
759 { 1500 {
760 memlen *= 2; 1501 namesalloc *= 2;
761 X_LOCK (wrklock);
762 req->ptr2 = names = realloc (names, memlen); 1502 req->ptr2 = names = realloc (names, namesalloc);
763 X_UNLOCK (wrklock);
764 1503
765 if (!names) 1504 if (!names)
1505 break;
1506 }
1507
1508 memcpy (names + namesoffs, name, len);
1509
1510 if (dents)
1511 {
1512 struct eio_dirent *ent;
1513
1514 if (ecb_expect_false (dentoffs == dentalloc))
1515 {
1516 dentalloc *= 2;
1517 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1518
1519 if (!dents)
766 break; 1520 break;
767 } 1521 }
768 1522
769 memcpy (names + memofs, name, len); 1523 ent = dents + dentoffs;
770 memofs += len; 1524
1525 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1526 ent->namelen = len - 1;
1527 ent->inode = D_INO (entp);
1528
1529 inode_bits |= ent->inode;
1530
1531 switch (D_TYPE (entp))
1532 {
1533 default:
1534 ent->type = EIO_DT_UNKNOWN;
1535 flags |= EIO_READDIR_FOUND_UNKNOWN;
1536 break;
1537
1538 #ifdef DT_FIFO
1539 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1540 #endif
1541 #ifdef DT_CHR
1542 case DT_CHR: ent->type = EIO_DT_CHR; break;
1543 #endif
1544 #ifdef DT_MPC
1545 case DT_MPC: ent->type = EIO_DT_MPC; break;
1546 #endif
1547 #ifdef DT_DIR
1548 case DT_DIR: ent->type = EIO_DT_DIR; break;
1549 #endif
1550 #ifdef DT_NAM
1551 case DT_NAM: ent->type = EIO_DT_NAM; break;
1552 #endif
1553 #ifdef DT_BLK
1554 case DT_BLK: ent->type = EIO_DT_BLK; break;
1555 #endif
1556 #ifdef DT_MPB
1557 case DT_MPB: ent->type = EIO_DT_MPB; break;
1558 #endif
1559 #ifdef DT_REG
1560 case DT_REG: ent->type = EIO_DT_REG; break;
1561 #endif
1562 #ifdef DT_NWK
1563 case DT_NWK: ent->type = EIO_DT_NWK; break;
1564 #endif
1565 #ifdef DT_CMP
1566 case DT_CMP: ent->type = EIO_DT_CMP; break;
1567 #endif
1568 #ifdef DT_LNK
1569 case DT_LNK: ent->type = EIO_DT_LNK; break;
1570 #endif
1571 #ifdef DT_SOCK
1572 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1573 #endif
1574 #ifdef DT_DOOR
1575 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1576 #endif
1577 #ifdef DT_WHT
1578 case DT_WHT: ent->type = EIO_DT_WHT; break;
1579 #endif
1580 }
1581
1582 ent->score = 7;
1583
1584 if (flags & EIO_READDIR_DIRS_FIRST)
1585 {
1586 if (ent->type == EIO_DT_UNKNOWN)
1587 {
1588 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1589 ent->score = 1;
1590 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1591 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1592 }
1593 else if (ent->type == EIO_DT_DIR)
1594 ent->score = 0;
1595 }
771 } 1596 }
1597
1598 namesoffs += len;
1599 ++dentoffs;
772 } 1600 }
773 1601
774 if (errno) 1602 if (EIO_CANCELLED (req))
775 res = -1; 1603 {
1604 errno = ECANCELED;
1605 break;
1606 }
1607
1608#ifdef _WIN32
1609 if (!FindNextFile (dirp, &entp))
1610 {
1611 FindClose (dirp);
1612 dirp = 0;
1613 }
1614#endif
1615 }
1616}
1617
1618/*****************************************************************************/
1619/* working directory stuff */
1620/* various deficiencies in the posix 2008 api force us to */
1621/* keep the absolute path in string form at all times */
1622/* fuck yeah. */
1623
1624#if !HAVE_AT
1625
1626/* a bit like realpath, but usually faster because it doesn'T have to return */
1627/* an absolute or canonical path */
1628static const char *
1629wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1630{
1631 if (!wd || *path == '/')
1632 return path;
1633
1634 if (path [0] == '.' && !path [1])
1635 return wd->str;
1636
776 1637 {
777 req->result = res; 1638 int l1 = wd->len;
1639 int l2 = strlen (path);
1640
1641 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1642
1643 memcpy (res, wd->str, l1);
1644 res [l1] = '/';
1645 memcpy (res + l1 + 1, path, l2 + 1);
1646
1647 return res;
1648 }
778} 1649}
1650
1651#endif
1652
1653static eio_wd
1654eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1655{
1656 int fd;
1657 eio_wd res;
1658 int len = eio__realpath (tmpbuf, wd, path);
1659
1660 if (len < 0)
1661 return EIO_INVALID_WD;
1662
1663#if HAVE_AT
1664 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1665
1666 if (fd < 0)
1667 return EIO_INVALID_WD;
1668#endif
1669
1670 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1671
1672#if HAVE_AT
1673 res->fd = fd;
1674#endif
1675
1676 res->len = len;
1677 memcpy (res->str, tmpbuf->ptr, len);
1678 res->str [len] = 0;
1679
1680 return res;
1681}
1682
1683eio_wd
1684eio_wd_open_sync (eio_wd wd, const char *path)
1685{
1686 struct etp_tmpbuf tmpbuf = { };
1687 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1688 free (tmpbuf.ptr);
1689
1690 return wd;
1691}
1692
1693void
1694eio_wd_close_sync (eio_wd wd)
1695{
1696 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1697 {
1698 #if HAVE_AT
1699 eio__close (wd->fd);
1700 #endif
1701 free (wd);
1702 }
1703}
1704
1705#if HAVE_AT
1706
1707static int
1708eio__renameat2 (int olddirfd, const char *oldpath, int newdirfd, const char *newpath, unsigned int flags)
1709{
1710#if HAVE_RENAMEAT2
1711 return syscall (SYS_renameat2, olddirfd, oldpath, newdirfd, newpath, flags);
1712#else
1713 if (flags)
1714 return EIO_ENOSYS ();
1715
1716 return renameat (olddirfd, oldpath, newdirfd, newpath);
1717#endif
1718}
1719
1720/* they forgot these */
1721
1722static int
1723eio__truncateat (int dirfd, const char *path, off_t length)
1724{
1725 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1726 int res;
1727
1728 if (fd < 0)
1729 return fd;
1730
1731 res = ftruncate (fd, length);
1732 silent_close (fd);
1733 return res;
1734}
1735
1736static int
1737eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1738{
1739 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1740 int res;
1741
1742 if (fd < 0)
1743 return fd;
1744
1745 res = fstatvfs (fd, buf);
1746 silent_close (fd);
1747 return res;
1748}
1749
1750#endif
779 1751
780/*****************************************************************************/ 1752/*****************************************************************************/
781 1753
782#define ALLOC(len) \ 1754#define ALLOC(len) \
783 if (!req->ptr2) \ 1755 if (!req->ptr2) \
784 { \ 1756 { \
785 X_LOCK (wrklock); \ 1757 X_LOCK (EIO_POOL->wrklock); \
786 req->flags |= EIO_FLAG_PTR2_FREE; \ 1758 req->flags |= EIO_FLAG_PTR2_FREE; \
787 X_UNLOCK (wrklock); \ 1759 X_UNLOCK (EIO_POOL->wrklock); \
788 req->ptr2 = malloc (len); \ 1760 req->ptr2 = malloc (len); \
789 if (!req->ptr2) \ 1761 if (!req->ptr2) \
790 { \ 1762 { \
791 errno = ENOMEM; \ 1763 errno = ENOMEM; \
792 req->result = -1; \ 1764 req->result = -1; \
793 break; \ 1765 goto alloc_fail; \
794 } \ 1766 } \
795 } 1767 }
796 1768
797X_THREAD_PROC (eio_proc) 1769/*****************************************************************************/
798{
799 eio_req *req;
800 struct timespec ts;
801 worker *self = (worker *)thr_arg;
802 1770
803 /* try to distribute timeouts somewhat randomly */ 1771static void
804 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1772eio__slurp (int fd, eio_req *req)
1773{
1774 req->result = fd;
805 1775
806 for (;;) 1776 if (fd < 0)
1777 return;
1778
1779 if (req->offs < 0 || !req->size) /* do we need the size? */
807 { 1780 {
808 ts.tv_sec = time (0) + IDLE_TIMEOUT; 1781 off_t size = lseek (fd, 0, SEEK_END);
809 1782
810 X_LOCK (reqlock); 1783 if (req->offs < 0)
1784 req->offs += size;
811 1785
812 for (;;) 1786 if (!req->size)
813 { 1787 req->size = size - req->offs;
814 self->req = req = reqq_shift (&req_queue);
815
816 if (req)
817 break;
818
819 ++idle;
820
821 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
822 {
823 if (idle > max_idle)
824 {
825 --idle;
826 X_UNLOCK (reqlock);
827 X_LOCK (wrklock);
828 --started;
829 X_UNLOCK (wrklock);
830 goto quit;
831 }
832
833 /* we are allowed to idle, so do so without any timeout */
834 X_COND_WAIT (reqwait, reqlock);
835 ts.tv_sec = time (0) + IDLE_TIMEOUT;
836 }
837
838 --idle;
839 }
840
841 --nready;
842
843 X_UNLOCK (reqlock);
844
845 errno = 0; /* strictly unnecessary */
846
847 if (!EIO_CANCELLED (req))
848 switch (req->type)
849 {
850 case EIO_READ: ALLOC (req->size);
851 req->result = req->offs >= 0
852 ? pread (req->int1, req->ptr2, req->size, req->offs)
853 : read (req->int1, req->ptr2, req->size); break;
854 case EIO_WRITE: req->result = req->offs >= 0
855 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
856 : write (req->int1, req->ptr2, req->size); break;
857
858 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
859 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
860
861 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
862 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
863 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
864 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
865 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
866 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
867
868 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
869 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
870 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
871 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
872 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
873 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
874
875 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
876 case EIO_CLOSE: req->result = close (req->int1); break;
877 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
878 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
879 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
880 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
881 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
882 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
883 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
884 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
885
886 case EIO_READLINK: ALLOC (NAME_MAX);
887 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
888
889 case EIO_SYNC: req->result = 0; sync (); break;
890 case EIO_FSYNC: req->result = fsync (req->int1); break;
891 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
892
893 case EIO_READDIR: eio__scandir (req, self); break;
894
895 case EIO_BUSY:
896#ifdef _WIN32
897 Sleep (req->nv1 * 1000.);
898#else
899 {
900 struct timeval tv;
901
902 tv.tv_sec = req->nv1;
903 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
904
905 req->result = select (0, 0, 0, 0, &tv);
906 }
907#endif
908 break;
909
910 case EIO_UTIME:
911 case EIO_FUTIME:
912 {
913 struct timeval tv[2];
914 struct timeval *times;
915
916 if (req->nv1 != -1. || req->nv2 != -1.)
917 {
918 tv[0].tv_sec = req->nv1;
919 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
920 tv[1].tv_sec = req->nv2;
921 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
922
923 times = tv;
924 }
925 else
926 times = 0;
927
928
929 req->result = req->type == EIO_FUTIME
930 ? futimes (req->int1, times)
931 : utimes (req->ptr1, times);
932 }
933
934 case EIO_GROUP:
935 case EIO_NOP:
936 req->result = 0;
937 break;
938
939 case EIO_QUIT:
940 goto quit;
941
942 default:
943 req->result = -1;
944 break;
945 }
946
947 req->errorno = errno;
948
949 X_LOCK (reslock);
950
951 ++npending;
952
953 if (!reqq_push (&res_queue, req) && want_poll_cb)
954 want_poll_cb ();
955
956 self->req = 0;
957 worker_clear (self);
958
959 X_UNLOCK (reslock);
960 } 1788 }
961 1789
962quit: 1790 ALLOC (req->size);
963 X_LOCK (wrklock); 1791 req->result = pread (fd, req->ptr2, req->size, req->offs);
964 worker_free (self);
965 X_UNLOCK (wrklock);
966 1792
967 return 0; 1793 silent_close (fd);
968}
969 1794
970/*****************************************************************************/ 1795alloc_fail:
971 1796 ;
972static void eio_atfork_prepare (void)
973{
974 X_LOCK (wrklock);
975 X_LOCK (reqlock);
976 X_LOCK (reslock);
977#if !HAVE_PREADWRITE
978 X_LOCK (preadwritelock);
979#endif
980#if !HAVE_READDIR_R
981 X_LOCK (readdirlock);
982#endif
983} 1797}
984 1798
985static void eio_atfork_parent (void) 1799int ecb_cold
986{
987#if !HAVE_READDIR_R
988 X_UNLOCK (readdirlock);
989#endif
990#if !HAVE_PREADWRITE
991 X_UNLOCK (preadwritelock);
992#endif
993 X_UNLOCK (reslock);
994 X_UNLOCK (reqlock);
995 X_UNLOCK (wrklock);
996}
997
998static void eio_atfork_child (void)
999{
1000 eio_req *prv;
1001
1002 while (prv = reqq_shift (&req_queue))
1003 eio_destroy (prv);
1004
1005 while (prv = reqq_shift (&res_queue))
1006 eio_destroy (prv);
1007
1008 while (wrk_first.next != &wrk_first)
1009 {
1010 worker *wrk = wrk_first.next;
1011
1012 if (wrk->req)
1013 eio_destroy (wrk->req);
1014
1015 worker_clear (wrk);
1016 worker_free (wrk);
1017 }
1018
1019 started = 0;
1020 idle = 0;
1021 nreqs = 0;
1022 nready = 0;
1023 npending = 0;
1024
1025 eio_atfork_parent ();
1026}
1027
1028int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1800eio_init (void (*want_poll)(void), void (*done_poll)(void))
1029{ 1801{
1030 want_poll_cb = want_poll; 1802 eio_want_poll_cb = want_poll;
1031 done_poll_cb = done_poll; 1803 eio_done_poll_cb = done_poll;
1032 1804
1033#ifdef _WIN32 1805 return etp_init (EIO_POOL, 0, 0, 0);
1034 X_MUTEX_CHECK (wrklock);
1035 X_MUTEX_CHECK (reslock);
1036 X_MUTEX_CHECK (reqlock);
1037 X_MUTEX_CHECK (reqwait);
1038 X_MUTEX_CHECK (preadwritelock);
1039 X_MUTEX_CHECK (readdirlock);
1040
1041 X_COND_CHECK (reqwait);
1042#endif
1043
1044 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1045} 1806}
1046 1807
1808ecb_inline void
1047static void eio_api_destroy (eio_req *req) 1809eio_api_destroy (eio_req *req)
1048{ 1810{
1049 free (req); 1811 free (req);
1050} 1812}
1051 1813
1052#define REQ(rtype) \ 1814#define REQ(rtype) \
1053 eio_req *req; \ 1815 eio_req *req; \
1054 \ 1816 \
1055 req = (eio_req *)calloc (1, sizeof *req); \ 1817 req = (eio_req *)calloc (1, sizeof *req); \
1056 if (!req) \ 1818 if (!req) \
1057 return 0; \ 1819 return 0; \
… …
1071 { \ 1833 { \
1072 eio_api_destroy (req); \ 1834 eio_api_destroy (req); \
1073 return 0; \ 1835 return 0; \
1074 } 1836 }
1075 1837
1838#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1839
1840static void
1841eio_execute (etp_worker *self, eio_req *req)
1842{
1843#if HAVE_AT
1844 int dirfd;
1845#else
1846 const char *path;
1847#endif
1848
1849 if (ecb_expect_false (EIO_CANCELLED (req)))
1850 {
1851 req->result = -1;
1852 req->errorno = ECANCELED;
1853 return;
1854 }
1855
1856 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1857 {
1858 req->result = -1;
1859 req->errorno = ENOENT;
1860 return;
1861 }
1862
1863 if (req->type >= EIO_OPEN)
1864 {
1865 #if HAVE_AT
1866 dirfd = WD2FD (req->wd);
1867 #else
1868 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1869 #endif
1870 }
1871
1872 switch (req->type)
1873 {
1874 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1875 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1876 break;
1877 case EIO_WD_CLOSE: req->result = 0;
1878 eio_wd_close_sync (req->wd); break;
1879
1880 case EIO_SEEK: eio__lseek (req); break;
1881 case EIO_READ: ALLOC (req->size);
1882 req->result = req->offs >= 0
1883 ? pread (req->int1, req->ptr2, req->size, req->offs)
1884 : read (req->int1, req->ptr2, req->size); break;
1885 case EIO_WRITE: req->result = req->offs >= 0
1886 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1887 : write (req->int1, req->ptr2, req->size); break;
1888
1889 case EIO_FCNTL: req->result = fcntl (req->int1, (int) req->int2, req->ptr2); break;
1890 case EIO_IOCTL: req->result = ioctl (req->int1, (unsigned long)req->int2, req->ptr2); break;
1891
1892 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1893 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1894
1895#if HAVE_AT
1896
1897 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1898 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1899 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1900 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1901 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1902 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1903 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1904 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1905 case EIO_SLURP: eio__slurp ( openat (dirfd, req->ptr1, O_RDONLY | O_CLOEXEC), req); break;
1906
1907 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1908 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1909 req->result = req->wd && SINGLEDOT (req->ptr1)
1910 ? rmdir (req->wd->str)
1911 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1912 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1913 case EIO_RENAME: req->result = eio__renameat2 (
1914 dirfd,
1915 /* complications arise because "." cannot be renamed, so we might have to expand */
1916 req->wd && SINGLEDOT (req->ptr1) ? req->wd->str : req->ptr1,
1917 WD2FD ((eio_wd)req->int3),
1918 req->ptr2,
1919 req->int2
1920 );
1921 break;
1922 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1923 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1924 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1925 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1926 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1927 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1928 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, EIO_PATH_MAX);
1929 if (req->result == EIO_PATH_MAX)
1930 {
1931 req->result = -1;
1932 errno = ENAMETOOLONG;
1933 }
1934 break;
1935 case EIO_UTIME:
1936 case EIO_FUTIME:
1937 {
1938 struct timespec ts[2];
1939 struct timespec *times;
1940
1941 if (req->nv1 != -1. || req->nv2 != -1.)
1942 {
1943 ts[0].tv_sec = req->nv1;
1944 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1945 ts[1].tv_sec = req->nv2;
1946 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1947
1948 times = ts;
1949 }
1950 else
1951 times = 0;
1952
1953 req->result = req->type == EIO_FUTIME
1954 ? futimens (req->int1, times)
1955 : utimensat (dirfd, req->ptr1, times, 0);
1956 }
1957 break;
1958
1959#else
1960
1961 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1962 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1963 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1964 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1965 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1966 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1967 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1968 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1969 case EIO_SLURP: eio__slurp ( open (path , O_RDONLY | O_CLOEXEC), req); break;
1970
1971 case EIO_UNLINK: req->result = unlink (path ); break;
1972 case EIO_RMDIR: req->result = rmdir (path ); break;
1973 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1974 case EIO_RENAME: req->result = req->int2 ? EIO_ENOSYS () : rename (path, req->ptr2); break;
1975 case EIO_LINK: req->result = link (path , req->ptr2); break;
1976 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1977 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1978 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1979 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1980 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1981 req->result = readlink (path, req->ptr2, EIO_PATH_MAX);
1982 if (req->result == EIO_PATH_MAX)
1983 {
1984 req->result = -1;
1985 errno = ENAMETOOLONG;
1986 }
1987 break;
1988
1989 case EIO_UTIME:
1990 case EIO_FUTIME:
1991 {
1992 struct timeval tv[2];
1993 struct timeval *times;
1994
1995 if (req->nv1 != -1. || req->nv2 != -1.)
1996 {
1997 tv[0].tv_sec = req->nv1;
1998 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1999 tv[1].tv_sec = req->nv2;
2000 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2001
2002 times = tv;
2003 }
2004 else
2005 times = 0;
2006
2007 req->result = req->type == EIO_FUTIME
2008 ? futimes (req->int1, times)
2009 : utimes (req->ptr1, times);
2010 }
2011 break;
2012
2013#endif
2014
2015 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2016 {
2017 ALLOC (req->result);
2018 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2019 }
2020 break;
2021
2022 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2023 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
2024
2025 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2026 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2027
2028 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
2029 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
2030 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
2031
2032 case EIO_CLOSE: req->result = eio__close (req->int1); break;
2033 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
2034 case EIO_SYNC: req->result = 0; sync (); break;
2035 case EIO_FSYNC: req->result = fsync (req->int1); break;
2036 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2037 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2038 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2039 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2040 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2041 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2042 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2043 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2044
2045 case EIO_READDIR: eio__scandir (req, self); break;
2046
2047 case EIO_BUSY:
2048#ifdef _WIN32
2049 Sleep (req->nv1 * 1e3);
2050#else
2051 {
2052 struct timeval tv;
2053
2054 tv.tv_sec = req->nv1;
2055 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
2056
2057 req->result = select (0, 0, 0, 0, &tv);
2058 }
2059#endif
2060 break;
2061
2062#if 0
2063 case EIO_GROUP:
2064 abort (); /* handled in eio_request */
2065#endif
2066
2067 case EIO_NOP:
2068 req->result = 0;
2069 break;
2070
2071 case EIO_CUSTOM:
2072 req->feed (req);
2073 break;
2074
2075 default:
2076 req->result = EIO_ENOSYS ();
2077 break;
2078 }
2079
2080alloc_fail:
2081 req->errorno = errno;
2082}
2083
2084#ifndef EIO_NO_WRAPPERS
2085
2086eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2087{
2088 REQ (EIO_WD_OPEN); PATH; SEND;
2089}
2090
2091eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2092{
2093 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2094}
2095
1076eio_req *eio_nop (int pri, eio_cb cb, void *data) 2096eio_req *eio_nop (int pri, eio_cb cb, void *data)
1077{ 2097{
1078 REQ (EIO_NOP); SEND; 2098 REQ (EIO_NOP); SEND;
1079} 2099}
1080 2100
… …
1091eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) 2111eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1092{ 2112{
1093 REQ (EIO_FSYNC); req->int1 = fd; SEND; 2113 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1094} 2114}
1095 2115
2116eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2117{
2118 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2119}
2120
1096eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2121eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1097{ 2122{
1098 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2123 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1099} 2124}
1100 2125
2126eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2127{
2128 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2129}
2130
2131eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2132{
2133 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2134}
2135
2136eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2137{
2138 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2139}
2140
2141eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2142{
2143 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2144}
2145
2146eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2147{
2148 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2149}
2150
2151eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2152{
2153 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2154}
2155
1101eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2156eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1102{ 2157{
1103 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2158 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1104} 2159}
1105 2160
1106eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2161eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1107{ 2162{
1108 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2163 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1109} 2164}
1110 2165
2166eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2167{
2168 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2169}
2170
1111eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2171eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1112{ 2172{
1113 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2173 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1114} 2174}
1115 2175
1116eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2176eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1117{ 2177{
1118 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2178 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1119} 2179}
1120 2180
2181eio_req *eio_fcntl (int fd, int cmd, void *arg, int pri, eio_cb cb, void *data)
2182{
2183 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = cmd; req->ptr2 = arg; SEND;
2184}
2185
2186eio_req *eio_ioctl (int fd, unsigned long request, void *buf, int pri, eio_cb cb, void *data)
2187{
2188 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = request; req->ptr2 = buf; SEND;
2189}
2190
1121eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2191eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1122{ 2192{
1123 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2193 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1124} 2194}
1125 2195
2196eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2197{
2198 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2199}
2200
1126eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2201eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1127{ 2202{
1128 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2203 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1129} 2204}
1130 2205
… …
1136eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2211eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1137{ 2212{
1138 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2213 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1139} 2214}
1140 2215
1141eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2216eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1142{ 2217{
1143 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2218 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1144} 2219}
1145 2220
1146eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2221eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1166eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2241eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1167{ 2242{
1168 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2243 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1169} 2244}
1170 2245
1171eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2246eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1172{ 2247{
1173 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2248 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1174} 2249}
1175 2250
1176eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2251eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1192eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2267eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1193{ 2268{
1194 return eio__1path (EIO_READLINK, path, pri, cb, data); 2269 return eio__1path (EIO_READLINK, path, pri, cb, data);
1195} 2270}
1196 2271
2272eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2273{
2274 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2275}
2276
1197eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2277eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1198{ 2278{
1199 return eio__1path (EIO_STAT, path, pri, cb, data); 2279 return eio__1path (EIO_STAT, path, pri, cb, data);
1200} 2280}
1201 2281
1202eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2282eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1203{ 2283{
1204 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2284 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1205} 2285}
1206 2286
2287eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2288{
2289 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2290}
2291
1207eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2292eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1208{ 2293{
1209 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2294 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1210} 2295}
1211 2296
1212eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2297eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1213{ 2298{
1214 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2299 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1215} 2300}
1216 2301
1217eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2302eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1218{ 2303{
1219 return eio__1path (EIO_READDIR, path, pri, cb, data); 2304 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1220} 2305}
1221 2306
1222eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2307eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1223{ 2308{
1224 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 2309 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1225} 2310}
1226 2311
1227static eio_req * 2312static eio_req *
1228eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2313eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1229{ 2314{
… …
1253eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2338eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1254{ 2339{
1255 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2340 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1256} 2341}
1257 2342
2343eio_req *eio_slurp (const char *path, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2344{
2345 REQ (EIO_SLURP); PATH; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2346}
2347
2348eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2349{
2350 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2351}
2352
2353#endif
2354
1258eio_req *eio_grp (eio_cb cb, void *data) 2355eio_req *eio_grp (eio_cb cb, void *data)
1259{ 2356{
1260 const int pri = EIO_PRI_MAX; 2357 const int pri = EIO_PRI_MAX;
1261 2358
1262 REQ (EIO_GROUP); SEND; 2359 REQ (EIO_GROUP); SEND;
… …
1267#undef SEND 2364#undef SEND
1268 2365
1269/*****************************************************************************/ 2366/*****************************************************************************/
1270/* grp functions */ 2367/* grp functions */
1271 2368
2369void
1272void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2370eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1273{ 2371{
1274 grp->int2 = limit; 2372 grp->int2 = limit;
1275 grp->feed = feed; 2373 grp->feed = feed;
1276 2374
1277 grp_try_feed (grp); 2375 grp_try_feed (grp);
1278} 2376}
1279 2377
2378void
1280void eio_grp_limit (eio_req *grp, int limit) 2379eio_grp_limit (eio_req *grp, int limit)
1281{ 2380{
1282 grp->int2 = limit; 2381 grp->int2 = limit;
1283 2382
1284 grp_try_feed (grp); 2383 grp_try_feed (grp);
1285} 2384}
1286 2385
2386void
1287void eio_grp_add (eio_req *grp, eio_req *req) 2387eio_grp_add (eio_req *grp, eio_req *req)
1288{ 2388{
1289 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2389 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2390
2391 grp->flags |= ETP_FLAG_GROUPADD;
1290 2392
1291 ++grp->size; 2393 ++grp->size;
1292 req->grp = grp; 2394 req->grp = grp;
1293 2395
1294 req->grp_prev = 0; 2396 req->grp_prev = 0;
… …
1301} 2403}
1302 2404
1303/*****************************************************************************/ 2405/*****************************************************************************/
1304/* misc garbage */ 2406/* misc garbage */
1305 2407
2408eio_ssize_t
1306ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2409eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1307{ 2410{
1308 worker wrk;
1309
1310 wrk.dbuf = 0;
1311
1312 eio__sendfile (ofd, ifd, offset, count, &wrk); 2411 return eio__sendfile (ofd, ifd, offset, count);
1313
1314 if (wrk.dbuf)
1315 free (wrk.dbuf);
1316} 2412}
1317 2413

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines